SAFEVID aims to develop and evaluate a privacy-preserving artificial intelligence (AI) approach that can analyse infant movement videos held in multiple Trusted Research Environments (TREs), supporting earlier identification of neurological conditions without sharing sensitive patient data.
Many neurological conditions in infants are difficult to diagnose early, even though earlier intervention can improve outcomes. AI systems need access to large and diverse video datasets, but these are often held by different organisations and cannot easily be shared.
SAFEVID (Spasms Analysis using Federated Learning from Videos Across Multiple TREs) addresses this challenge by using federated learning, which allows AI models to learn from data across multiple TREs while keeping the data securely in place.
Earlier identification of neurological conditions could help infants and families access support sooner and reduce inequalities in access to specialist assessment. By analysing smartphone videos uploaded by parents and carers, SAFEVID could make screening more accessible and help identify children who may benefit from further clinical review.
The SAFEVID project is led by the University of Glasgow and delivered in partnership with the Health Informatics Centre (HIC), University of Dundee, the NHS West of Scotland Innovation Hub, and vCreate, combining expertise in TREs, data governance, secure video sharing and AI-enabled healthcare research.
SAFEVID is exploring two TREvolution capabilities. SACRO-ML helps assess disclosure risks before machine learning models are shared between organisations, while K8TRE provides a secure and scalable TRE platform that supports collaborative research. Together, these technologies enable AI models to be trained and evaluated across multiple organisations while maintaining strong privacy and governance standards.
What does SAFEVID aim to achieve
By the end of the project, SAFEVID aims to demonstrate how privacy-preserving AI can support secure analysis of infant movement videos across multiple TREs. The goals include:
- Assess federated learning – Evaluating federated learning across TREs hosted by HIC and the NHS West of Scotland Innovation Hub.
- Improving AI for neurological screening – Developing and validating AI models that identify early signs of neurological conditions from infant movement videos.
- Maintaining privacy – Assessing privacy risks associated with sharing machine learning models and evaluating safeguards using SACRO-ML.
- Install, test and examine federated learning infrastructure – Deploying and testing a secure federated learning infrastructure using K8TRE and comparing model performance across different datasets and TREs.
- Creating guidance – Producing practical guidelines for conducting federated AI research using sensitive healthcare video data.
- Embedding public and patient perspectives – Engaging patients, families and public contributors to ensure the technology is developed responsibly and transparently.
Progress so far
Within the first six months of delivery, the SAFEVID team has:
- Established regular collaboration meetings between all project partners to coordinate research, governance and technical activities.
- Obtained a favourable ethical opinion from the NHS Research Ethics Committee (REC).
- Developed an initial federated learning prototype
- Conducted research on using uncertainty estimates to improve AI performance in federated learning.
- Had two research papers accepted for presentation at the 4th International Conference on Federated Learning Technologies and Applications (FLTA 2026).
- Launched Patient and Public Involvement and Engagement (PPIE) activities, including an online questionnaire and recruitment for an in-person workshop planned for Autumn 2026.
- Started development of a SATRE-compliant TRE at the NHS West of Scotland Innovation Hub.
- Begun integrating TREvolution technologies, including K8TRE and SACRO-ML, into the project workflow.
Over the remaining six months, the team will complete the TRE deployment, implement and evaluate federated learning across partner organisations, and assess disclosure risks using SACRO-ML. The project will also expand public engagement activities, deliver its planned workshop, and produce guidance to support future federated AI research using sensitive healthcare video data.
Relevance to TREvolution
SAFEVID supports TREvolution’s goal of enabling secure and trustworthy research across connected TREs. The project will provide practical evidence, technical solutions and governance recommendations for privacy-preserving AI, helping researchers, TRE providers and healthcare organisations adopt federated approaches safely and effectively.
About the DARE UK Research Exemplars
The DARE UK Research Exemplars are short, impactful research projects demonstrating how emerging Trusted Research Environment (TRE) capabilities can support real-world sensitive data research across the UK. Working across multiple TREs and research domains, these projects bring together researchers, TRE operators, public contributors and other partners to apply and evaluate innovations developed through TREvolution. Their role is to test new approaches in live research settings, generating practical evidence on what works well, what challenges remain, and how sensitive data research can be strengthened for the public good.
